JP6355908B2 - Symbol recognition device and vehicle sign recognition device - Google Patents

Description

  The present invention relates to an apparatus for recognizing a symbol in an image, and more particularly to a symbol recognition apparatus suitable for recognizing a number included in a road speed sign. The present invention also relates to a vehicular sign recognition device provided with such a symbol recognition device.

  In recent years, the increase in the number of traffic accidents has become serious due to the increase in the number of automobiles used. The main causes of traffic accidents are those caused by drivers such as late discovery, misjudgment, and operational mistakes. In order to prevent a traffic accident, it is possible to prevent a driver's operation mistake or a judgment error by making a decision with an automobile and assisting the driver's operation. In order to enable such a determination in a car, it is necessary to perform white line recognition, pedestrian recognition, vehicle recognition, road sign recognition, etc. at a time with real-time characteristics that can be processed while the car is running.

  Speaking of road sign recognition, detection of road speed signs and numeric recognition are examples. Conventionally, symbols such as letters, numbers, and symbols are recognized by performing pattern matching after thinning the image (see, for example, Non-Patent Document 1).

Hideyuki Tamura, “Computer Image Processing”, Ohmsha, December 20, 2002, p.137-161

  Image thinning processing is a technique for software, and it is necessary to repeatedly read an image from an image memory. Therefore, it is difficult to implement this method in hardware such as an in-vehicle embedded system and perform real-time processing while the vehicle is running.

  In view of the above problems, an object of the present invention is to provide a symbol recognition technique suitable for hardware processing. Another object of the present invention is to provide a vehicular sign recognition apparatus to which such a symbol recognition technique is applied.

  A symbol recognition apparatus according to an aspect of the present invention receives an image memory that stores a binarized image to be recognized, and an image of each partial area of the recognition target area in the binarized image. A plurality of histogram calculation units for calculating the frequency distribution of pixels of a predetermined color in each line or each column and an image of each partial area of the recognition target area are input, and a predetermined length consisting of pixels of the predetermined color in the partial area A plurality of run length determination units for determining whether or not the above-described lines or columns exist, and the plurality of pieces of pixel information of the partial areas read by scanning the binary image stored in the image memory A control unit that inputs to the plurality of histogram calculation units and the plurality of run length determination units, a calculation result of the plurality of histogram calculation units, and a determination result of the plurality of run length determination units Based on, and a determination unit for determining symbols included in the binarized image.

  According to this, the binarized image stored in the image memory is read out by the control unit, and an image of an arbitrary partial region of the binarized image is input to each histogram calculation unit and each run length determination unit, and each histogram The calculation unit and each run length determination unit perform predetermined processing on each partial area independently and simultaneously in parallel. Based on the output results of each histogram calculation unit and each run length determination unit, the determination unit Symbols included in the binarized image are determined. Thus, the symbols included in the binarized image can be recognized by the control unit only scanning the binarized image once.

  The plurality of histogram calculation units divide the recognition target region into two parts in the left-right direction, first and second histogram calculation units to which an image of each partial region is input, and the recognition target region is divided into three in the vertical direction. Third, fourth, and fifth histogram calculators to which images of each partial area are input; a sixth histogram calculator to which an image of a partial area at the center in the horizontal direction of the recognition target area; And a seventh histogram calculation unit to which an image of a partial region at the center in the vertical direction of the recognition target region may be included. In addition, the first, second, third, and fourth images in which the plurality of run length determination units are inputted with the partial areas obtained by dividing the partial area in the vertical direction center of the recognition target area into four parts in the vertical and horizontal directions are input. The run length determination unit may be included. The determination unit includes a maximum value of the histogram calculated by the first to fifth histogram calculation units, a minimum value of the histogram calculated by the sixth and seventh histogram calculation units, and the first to You may make it determine the symbol contained in the said binarized image based on the determination result of a 4th run length determination part.

  According to this, numbers, particularly numbers included in road speed signs, can be recognized at high speed and with high accuracy.

  The plurality of run length determination units include a fifth run length determination unit, and the fifth run length determination unit receives an image of a partial region in the center in the vertical direction of the recognition target region, and the partial region It is determined whether or not there are a predetermined number or more of black and white columns having a predetermined length or longer, and the determination unit takes into account the determination result of the fifth run length determination unit, and the recognition target It may be determined whether or not the symbol included in the region is the number “0”.

  According to this, the recognition accuracy of the number “0” is improved.

  In addition, the plurality of run length determination units may include sixth and seventh run length determination units to which an image of each partial region obtained by dividing the recognition target region into two in the left-right direction, and the recognition target region in the vertical direction. And the eighth and ninth run length determination units to which the images of the upper and lower partial regions divided into three are input, and the determination unit determines the determination results of the sixth to ninth run length determination units In consideration of this, it may be determined whether or not the symbol included in the recognition target area is the number “0”.

  According to this, the recognition accuracy of the number “0” is further improved.

  Alternatively, the plurality of run-length determination units include sixth, seventh, and eighth run-length determination units to which images of partial regions obtained by dividing the recognition target region into three in the vertical direction are input. The unit may determine whether or not the symbol included in the recognition target area is the number “8” in consideration of the determination results of the sixth to eighth run length determination units.

  According to this, the recognition accuracy of the number “8” is improved.

  The symbol recognition apparatus may include a color ratio calculation unit that receives an image of the recognition target area and calculates a ratio of pixels of a predetermined color in the entire recognition target area, and the control unit includes the image The pixel information of the partial area read by scanning the binarized image stored in the memory may be input to the color ratio calculation unit, and the determination unit may include the color ratio calculation unit. A symbol included in the recognition target area may be determined in consideration of a calculation result.

  According to this, the symbol recognition accuracy is improved by taking into account the proportion of the predetermined color in the entire recognition target area.

  The symbol recognition device may perform a recognition process in parallel on the recognition target area and each recognition target area in which the recognition target area is slightly shifted vertically and horizontally.

  According to this, even if the symbol is shifted from the predetermined position in the binarized image, the symbol included in the binarized image can be recognized correctly.

  A vehicle sign recognition device according to another aspect of the present invention detects a frame of a vehicle sign in the symbol recognition device and the binarized image stored in the image memory of the symbol recognition device. A frame detection unit, and the determination unit of the symbol recognition device determines whether the binarized image is a vehicle sign image based on a detection result of the frame detection unit. .

  According to this, the determination of the vehicle sign and the recognition of the speed limit can be performed on the binarized image stored in the image memory.

  Furthermore, the vehicle sign recognition apparatus may include a diagonal line determination unit that determines whether or not there is a diagonal line having a predetermined length or more composed of pixels of a predetermined color in an oblique partial region in the binarized image. The determination unit may determine whether the binarized image is an image of any of the signs indicating prohibition in consideration of the determination result of the oblique line determination unit. Good.

  According to this, it is possible to determine whether or not the binarized image stored in the image memory is a sign indicating prohibition, and further to recognize the type of sign indicating prohibition.

  According to the present invention, it is possible to recognize symbols included in a binarized image at high speed and with high accuracy using hardware. In addition, it is possible to realize a low-cost in-vehicle built-in system that can perform a vehicle sign recognition process in real time while the automobile is running.

1 is a configuration diagram of a main part of a symbol recognition apparatus according to a first embodiment of the present invention. The figure which shows the example of the binarized image of recognition object The figure which shows the example of the partial area | region of the recognition object area | region processed by each feature-value extraction part. The figure which shows the characteristic of the histogram of the image of each number from "0" to "8" The figure which shows the correspondence of the histogram of the image of the number "5", and each partial area The figure which shows the run length characteristic of the image of each number from "0" to "8" The block diagram of the principal part of the symbol recognition apparatus which concerns on the 2nd Embodiment of this invention. The figure which shows the feature of the run length in the partial area | region with the number "0" The figure which shows the characteristic of the run length in another several partial area | region of the number "0" The figure which shows the feature of run length in some partial areas of number "8" The block diagram of the principal part of the symbol recognition apparatus which concerns on the 3rd Embodiment of this invention. The figure which shows the example of the reference recognition object area, and the example of the recognition object area shifted up and down, right and left The block diagram of the principal part of the marker recognition apparatus for vehicles which concerns on the 4th Embodiment of this invention. The figure which shows the example of the partial area | region of the binary image processed with a frame detection part. The figure which shows the example of the partial area | region of the binarized image processed with a diagonal determination part, and some examples of the label | marker showing prohibition

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In addition, this invention is not limited to the following embodiment.

<< First Embodiment >>
FIG. 1 shows the configuration of the main part of a symbol recognition apparatus according to the first embodiment of the present invention. The symbol recognition apparatus 10 according to the present embodiment recognizes symbols included in a binary image having a predetermined size, and is particularly suitable for recognizing numbers included in road speed signs. .

  The symbol recognition device 10 includes an image memory 12, a control unit 14, eleven feature quantity extraction units 16_1 to 16_11, and a determination unit 18. Each of the feature amount extraction units 16_1 to 16_11 includes a histogram calculation unit 161 and a run length determination unit 162. For convenience, the feature quantity extraction units 16_3 to 16_10 are not shown in FIG. 1, but in the following description, the feature quantity extraction units 16_3 to 16_10 (not shown) may be referred to.

  The image memory 12 is a memory that stores a binarized image to be recognized by the symbol recognition device 10. The image memory 12 can be composed of a volatile memory such as SRAM (Static Random Access Memory) or DRAM (Dynamic Random Access Memory).

  FIG. 2 shows an example of a binarized image to be recognized. The image is obtained by binarizing a 640 × 480 pixel 8-bit grayscale original image taken in front of a camera attached to a car to obtain a road speed sign candidate region (hereinafter referred to as a “scan window”). Extracted. The size of the scan window SW is about 20 × 20 pixels to 50 × 50 pixels.

  As shown in FIG. 2, the road speed sign is represented by a design in which a two-digit number is surrounded by a circle. The symbol recognition device 10 recognizes the first digit and the tenth digit included in the binarized image. That is, the symbol recognizing device 10 recognizes numbers included in each of two rectangular areas (two rectangular areas surrounded by a broken line in FIG. 2, hereinafter referred to as “recognition target areas”) at a substantially central portion of the scan window SW. To do. A technique for extracting a scan window by binarizing a grayscale original image is described in detail in the specification of Japanese Patent Application No. 2012-285669 by the present inventors.

  Returning to FIG. 1, the control unit 14 scans the binary image stored in the image memory 12 and reads out the pixel information of the partial area, that is, the pixel value Ixy = {0, 1} and the coordinates of the pixel. Is input to the histogram calculation unit 161 and the run length determination unit 162 in each of the feature amount extraction units 16_1 to 16_11. The pixel information can be read out in units of several pixels or in units of one pixel. More specifically, a coordinate range in the scan window is set in each partial area. The control unit 18 detects that the coordinate data (x, y) has entered the coordinate range of each partial area during scanning of the binarized image stored in the image memory 12, and corresponds to the partial area. A control signal ACT is output to each of the histogram calculation unit 161 and the run length determination unit 162 in each feature amount extraction unit 16_1 to 16_11. For example, each histogram calculation unit 161 and each run length determination unit 162 fetches the pixel value Ixy and coordinate data (x, y) when the control signal ACT input to the histogram calculation unit 161 and the run length determination unit 162 are at the H level, and the control signal ACT is at the L level. In this case, the pixel value Ixy and the coordinate data (x, y) are not captured. Thereby, it is possible to input an image of an arbitrary partial area of the binarized image to each histogram calculation unit 161 and each run length determination unit 162 by scanning the binarized image stored in the image memory 12 once. it can. Each histogram calculation unit 161 and each run length determination unit 162 perform a predetermined process on the input image simultaneously and independently from each other.

  Note that the control unit 14 does not directly read the pixel value Ixy from the image memory, but the control unit 14 outputs a control signal for reading the pixel value to the image memory 12, and is read from the image memory 12 according to the control signal. The pixel value Ixy may be directly supplied to each feature amount extraction unit 16_1 to 16_11.

  FIG. 3 shows an example of a partial region of the recognition target region processed by each feature amount extraction unit 16_1 to 16_11. The images of the partial areas BL1 and BL2 obtained by dividing the recognition target area into two in the left-right direction, for example, equally divided into two, are input to the feature amount extraction units 16_1 and 16_2 and processed. The images of the partial areas BL3, BL4, and BL5 obtained by dividing the recognition target area into three in the vertical direction, for example, equally divided into three, are input to the feature amount extraction units 16_3, 16_4, and 16_5 and processed. The image of the partial region BL6 at the center in the left-right direction of the recognition target region is input to the feature amount extraction unit 16_6 and processed. The image of the partial area BL7 in the center in the vertical direction of the recognition target area is input to the feature amount extraction unit 16_7 and processed. The images of the partial areas BL8, BL9, BL10, and BL11 obtained by dividing the partial area at the center in the vertical direction of the recognition target area into four parts in the vertical and horizontal directions, for example, equally divided into four are input to the feature amount extraction units 16_8, 16_9, 16_10, and 16_11. To be processed.

  Returning to FIG. 1, the histogram calculation unit 161 receives an image of a partial area of the recognition target area, and calculates a frequency distribution of pixels of a predetermined color in each line or each column in the partial area. In the case of a binarized image of a normal road speed sign, the number is displayed in black on a white background, so the histogram calculation unit 161 calculates the frequency distribution of black pixels. On the other hand, in the case of a binarized image of an electric display road speed sign, the number may be represented in white on a black background. In this case, the histogram calculation unit 161 calculates the frequency distribution of white pixels.

  Since the calculation of the histogram can be performed by an addition operation and a comparison operation, the histogram calculation unit 161 can basically be configured using an addition circuit and a comparator.

  FIG. 4 shows the characteristics of the histogram of each numerical image from “0” to “8”. In FIG. 4, the frequency distribution of black pixels in each line in the recognition target area is next to each number (the right side for the number “0” and the left side for other numbers), and the frequency of black pixels in each column in the recognition target area. The distribution is shown below each number. As shown in FIG. 4, the histogram has different characteristics for each number.

  FIG. 5 shows the correspondence between the histogram of the image “5” and the partial areas BL1 to BL7. As shown in FIG. 5A, the histogram calculation unit 161 in the feature amount extraction units 16_1 and 16_2 calculates the maximum value of the histogram corresponding to the partial regions BL1 and BL2 in the histograms in the column direction of the numerical images, respectively. Output. The histogram calculation units 161 in the feature amount extraction units 16_3 to 16_5 each output the maximum value of the histogram corresponding to the partial regions BL3 to BL5 in the line direction histogram of the numerical image. On the other hand, as shown in FIG. 5B, the histogram calculation unit 161 in the feature amount extraction unit 16_6 outputs the minimum value of the histogram corresponding to the partial region BL6 in the histogram in the column direction of the numerical image. The histogram calculation unit 161 in the feature amount extraction unit 16_7 outputs the minimum value of the histogram corresponding to the partial region BL7 in the histogram in the line direction of the number image. The maximum value and the minimum value of the histogram can be defined as a ratio of the frequency of pixels of a predetermined color included in each line or each column to the width or height of the recognition target area. For example, in the case of a normal road speed sign, the predetermined color is black, and in the electric speed display road speed sign, the predetermined color is white. If the symbol to be recognized is other than a number included in the road speed sign, the predetermined color may be white or black as appropriate. From the characteristics of the binarized image, the maximum value of the black pixel histogram corresponds to the minimum value of the white pixel histogram, and the maximum value of the white pixel histogram corresponds to the minimum value of the black pixel histogram. To do.

  As described above, the maximum values of the histograms of the partial areas BL1 to BL5 and the minimum values of the histograms of the partial areas BL6 to BL7 have different characteristics for each number, and the slope of the number, positional deviation, and noise superposition It is hard to be influenced by such. For this reason, the maximum value and the minimum value of the histogram of each partial region can be effective feature amounts for recognizing symbols in the recognition target region.

  Returning to FIG. 1, the run length determination unit 162 receives an image of each partial area of the recognition target area, and determines whether or not there is a line or a column having a predetermined length or more composed of pixels of a predetermined color in the partial area. To do. As will be described later, since the number image has a feature in the continuity of pixels of a predetermined color in the line direction in a portion other than the line segment representing the number, the run length determination unit 162 has a continuity of pixels of the predetermined color in the line direction. Determine sex. In addition, from the characteristics of the font design of the numbers used for road speed signs, it is appropriate that the areas for which the continuity of pixels of a predetermined color is to be determined are the partial areas BL8 to BL11. Accordingly, the run length determination unit 162 in the feature amount extraction units 16_8 to 16_11 determines whether or not a white or black line having a predetermined length or more exists in the partial regions BL8 to BL11, respectively. The predetermined length is an arbitrary length such as 90% or 95% with respect to the width of each of the partial regions BL8 to BL11. The width of each of the partial areas BL8 to BL11 can be specified as a predetermined length. In this case, the run length determination unit 162 determines whether or not all white lines or all black lines exist in the partial areas BL8 to BL11. It will be. In the case of a binarized image of a normal road speed sign, the number is displayed in black on a white background, and therefore the run length calculation unit 162 determines the continuity of white pixels. On the other hand, in the case of a binarized image of an electric display road speed sign, the number may be represented in white on a black background. In this case, the run length determination unit 162 determines the continuity of black pixels.

  Note that since the run length can be processed by AND operation, OR operation, and addition, the run length determination unit 162 can be configured as a combinational circuit including an AND circuit, an OR circuit, and an adder.

  FIG. 6 shows the run length characteristics of the images of the numbers from 0 to 8. In the number “1”, white lines having a predetermined length or longer exist in the partial areas BL9 and BL11, in the number “2”, white lines having a predetermined length or longer exist in the partial areas BL8 and BL11, and in the number “3”, the partial area BL8. And BL10 have a white line of a predetermined length or more, the number “4” has a white line of a predetermined length or more in the partial area BL8, and the number “5” has a white line of a predetermined length or more in the partial areas BL9 and BL10. In the number “6”, a white line having a predetermined length or longer exists in the partial region BL9, in the number “7”, a white line having a predetermined length or longer exists in the partial region BL8, and in the numbers “0” and “8”. There is no white line longer than a predetermined length in any partial area. Thus, the pattern of the presence or absence of a white line having a predetermined length or more in the partial areas BL8 to BL11 is different for each number. However, although the numbers “0” and “8” have the same pattern for the presence or absence of a white line of a predetermined length or more, the number “0” is not used at the tens place in the road speed sign, No digits other than the number “0” are used. Therefore, the numbers “0” and “8” can be distinguished according to the position of the recognition target region in the scan window.

  In the case where the degree of number inclination due to the inclination of the road speed sign, the position deviation due to the deviation of the scan window, the superimposition of noise, etc. are large, the pattern of the presence or absence of white lines of a predetermined length or more in the partial areas BL8 to BL11 It's different from the pattern. Therefore, it is difficult to recognize numbers only by the continuity of pixels of a predetermined color in the partial areas BL8 to BL11. Therefore, the determination unit 18 comprehensively determines numbers together with the histogram feature amounts of the partial regions BL1 to BL7.

  Returning to FIG. 1, the determination unit 18 recognizes the recognition target region based on the calculation result of the histogram calculation unit 161 in the feature amount extraction units 16_1 to 16_7 and the determination result of the run length determination unit 162 in the feature amount extraction units 16_8 to 16_11. The number contained in is determined. For example, the determination unit 18 determines a number according to the following determination condition. Note that, under the following determination conditions, Hmax1 to Hmax5 are maximum values of the histogram output from the histogram calculation unit 161 in the feature amount extraction units 16_1 to 16_5, and Hmin6 and Hmin7 are histograms in the feature amount extraction units 16_6 and 16_7. RL1 to RL4 are determination results output from the run length determination unit 162 in the feature amount extraction units 16_8 to 16_11. RL1 to RL4 are “1” when a white line longer than a predetermined length is detected, “0” when no white line is detected, and “*” indicates don't care.

<Determination of the number “0”>
(RL1, RL2, RL3, RL4) = (0, 0, 0, 0)
-Hmax1, Hmax2, Hmax3, Hmax5 are 50% or more-Hmax1-Hmin6, Hmax2-Hmin6 are 30% or more-| Hmax1-Hmax2 | is less than 50%-| Hmax3-Hmax5 | is less than 30%-Hmin6, Hmin7 is 10 When the above condition is satisfied, the determination unit 18 determines that the number included in the recognition target area is “0”.

<Determination of the number “1”>
・ (RL1, RL2, RL3, RL4) = (0, *, 0, *)
-Hmax1 is 70% or more-Hmax3 is less than 70%-Hmax4, Hmax5 is less than 50% or (RL1, RL2, RL3, RL4) = (0, *, 0, *)
・ Hmax1 is 70% or more ・ Hmax3, Hmax4, Hmax5 is less than 70% ・ Hmax1-Hmin6 is 50% or more ・ Hmin6 is 20% or less
When the above conditions are satisfied, the determination unit 18 determines that the number included in the recognition target area is “1”.

<Determination of the number “2”>
・ (RL1, RL2, RL3, RL4) = (1, 0, *, *)
-Hmax1, Hmax2, Hmax4 are less than 70%-Hmax3 is 50% or more-Hmax5 is 70% or more-Hmax2-Hmin6 is less than 30% When the above conditions are satisfied, It is determined that it is “2”.

<Determination of the number “3”>
(RL1, RL2, RL3, RL4) = (1, 0, 1, 0)
-Hmax1 is less than 70%-Hmax2, Hmax3, Hmax5 are 50% or more When the above condition is satisfied, the determination unit 18 determines that the number included in the recognition target region is "3".

<Determination of the number “4”>
(RL1, RL2, RL3, RL4) = (*, 0, 0, 0)
・ Hmax1 is less than 50% ・ Hmax2 and Hmax5 are 70% or more ・ Hmax3 and Hmax4 are less than 70% or (RL1, RL2, RL3, RL4) = (*, 0, 0, 0)
-Hmax1, Hmax3, Hmax4 is less than 70%-Hmax2, Hmax5 is 70% or more-Hmax2-Hmin6 is 50% or more-Hmax2-Hmax1 is 30% or more or-(RL1, RL2, RL3, RL4) = (*, 0 , 0, 0)
-Hmax2 and Hmax5 are 70% or more-Hmax3 is less than 70%-Hmax2-Hmin6 is 50% or more-Hmax5-Hmin7 is 40% or more When the above condition is satisfied, the determination unit 18 determines that the number included in the recognition target region is It is determined that it is “4”.

<Determination of the number “5”>
(RL1, RL2, RL3, RL4) = (0, 1, 1, 0)
When Hmax1, Hmax2, Hmax3, Hmax4, and Hmax5 satisfy the condition of 50% or more, the determination unit 18 determines that the number included in the recognition target area is “5”.

<Determination of the number “6”>
(RL1, RL2, RL3, RL4) = (0, 1, 0, 0)
When Hmax1, Hmax4, and Hmax5 are 50% or more and Hmax3 is less than 70%, the determination unit 18 determines that the number included in the recognition target area is “6”.

<Determination of number “7”>
・ (RL1, RL2, RL3, RL4) = (1, 0, *, *)
・ Hmax1, Hmax2, Hmax4, Hmax5 is less than 70% ・ Hmax3 is 70% or more or (RL1, RL2, RL3, RL4) = (1, 0, *, *)
-Hmax3 is 70% or more-Hmax3-Hmin7 is 50% or more When the conditions are met, the determination unit 18 determines that the number included in the recognition target area is "7".

<Determination of number “8”>
(RL1, RL2, RL3, RL4) = (0, 0, 0, 0)
・ Hmax1, Hmax2, Hmax3, Hmax4, Hmax5 are 70% or more or (RL1, RL2, RL3, RL4) = (0, 0, 0, 0)
・ Hmax1, Hmax2, Hmax3, Hmax4, Hmax5 are 50% or more ・ Hmax1-Hmin6 is 20% or more ・ Hmax2-Hmin6 is 20% or more ・ | Hmax1-Hmax2 | is less than 30% ・ | Hmax3-Hmax5 | is less than 30% When the above conditions are satisfied, the determination unit 18 determines that the number included in the recognition target area is “8”.

  If all the above conditions are not satisfied, the determination unit 18 may determine that the binarized image to be recognized is not a road speed sign image. Furthermore, in the road speed sign, only the number “0” appears in the recognition target area on the right side in the scan window (see FIG. 2). Therefore, the control unit 14 first scans the right recognition target area in the scan window, and the determination unit 18 determines whether the recognition target binary is a symbol “0” based on whether or not the symbol included in the recognition target area is the number “0”. It may be determined whether the converted image is an image of a road speed sign. When the symbol included in the right recognition target area in the scan window is the number “0”, the control unit 14 next scans the left recognition target area in the scan window. You may make it recognize the tens digit.

  As described above, according to the present embodiment, symbol recognition included in the binarized image stored in the image memory 12 can be performed at high speed and with high accuracy using hardware.

  Although the description of the determination condition of the number “9” is omitted, this is because the number “9” is not used for the road speed sign in Japan, and the number “9” is not unrecognizable. Based on the histogram and run length characteristics of the image of the number “9”, it is easy to recognize the number “9” like other numbers.

  In some cases, the speed is represented by a three-digit number in the road speed sign. In such a case, three recognition target areas may be set in the scan window so as to match each three-digit number. In addition, it is considered that the hundreds digit in the road speed sign cannot be other than “1”. Therefore, the control unit 14 first scans the leftmost recognition target area in the scan window, and the determination unit 18 determines whether the recognition target binary is a symbol “1” depending on whether or not the symbol included in the recognition target area is the number “1”. It may be determined whether the converted image is an image of a road speed sign. When the symbol included in the leftmost target area in the scan window is the number “1”, the control unit 14 next scans the middle or right end recognition target in the scan window, and the determination unit 18 You may make it recognize the number of the tenth place or the first place.

  Symbols other than numbers such as alphabets, hiragana and katakana can be recognized by the same method as described above. However, when recognizing symbols other than numbers, the partial areas for determining histograms and run lengths are not limited to the above-described partial areas BL1 to BL11, and include partial areas that reflect the characteristics of each symbol. It is desirable to adopt.

  Since the run length determination unit 162 in the feature quantity extraction units 16_1 to 16_7 and the histogram calculation unit 161 in the feature quantity extraction units 16_8 to 16_11 are not actually used, if the recognition is limited to numbers, these unused components are omitted. May be. Thereby, the circuit scale can be reduced. Further, the feature quantity extraction units 16_1 to 16_11 can share data and resources by optimizing, and the circuit scale can be reduced.

  Conversely, as in the present embodiment, an appropriate number of feature quantity extraction units 16 including a histogram calculation unit 161 and a run length determination unit 162 are arranged, and the control unit 14 controls the histogram calculation unit 161 in each feature quantity extraction unit 16. Further, by controlling the input of pixel information to the run length determination unit 162, the circuit scale increases, but versatility that can cope with recognition of various symbols other than numbers can be provided.

  When a plurality of partial regions overlap in the recognition target region, the entire circuit is obtained by sharing the feature amount extraction unit 16 so that one feature amount extraction unit 16 processes the plurality of partial regions. The scale can be reduced. However, since the circuit scale per feature quantity extraction unit 16 is very small, the influence of the increase in circuit scale is considered to be limited even if a large number of feature quantity extraction units 16 are arranged. Therefore, the feature amount extraction unit 16 may be shared when the circuit scale becomes a problem.

<< Second Embodiment >>
FIG. 7 shows the configuration of the main part of a symbol recognition apparatus according to the second embodiment of the present invention. The symbol recognition apparatus 10 according to the present embodiment adds a color ratio calculation unit 17 to the symbol recognition apparatus 10 according to the first embodiment, and further includes a run length determination unit 162 in the feature amount extraction units 16_1 to 16_5 and 16_7. Used to improve the recognition accuracy of numbers. Hereinafter, description of the same matters as in the first embodiment will be omitted, and points different from those in the first embodiment will be mainly described.

  The color ratio calculation unit 17 receives the image of the recognition target area directly from the image memory 12 or indirectly through the control unit 14 and calculates the ratio of pixels of a predetermined color in the recognition target area. The predetermined color is black in the case of a normal road speed sign, and is white in the case of an electric display road speed sign. The ratio of pixels of a predetermined color is, for example, 66% or less for the number “0”, 40% or less for the number “1”, 46% or less for the number “2”, 43% or less for the number “3”, and the number “4”. Is 50% or less, the number “5” is 54% or less, the number “6” is 48% or less, the number “7” is 44% or less, and the number “8” is 61% or less.

  The control unit 14 outputs a control signal ACT and controls input of pixel information to the color ratio calculation unit 17. For example, the color ratio calculation unit 17 takes in the pixel value Ixy and coordinate data (x, y) when the control signal ACT input to the H level is H level, and captures the pixel value Ixy and the coordinate data (x, y) when the control signal ACT is L level. The coordinate data (x, y) is not captured. As a result, the control unit 14 scans the binarized image stored in the image memory 12 once, and images of arbitrary partial areas of the binarized image are extracted from the feature amount extracting units 16_1 to 16_11 and the color ratio calculating unit 17. Can be entered. Then, each of the feature amount extraction units 16_1 to 16_11 and the color ratio calculation unit 17 performs predetermined processing on the input image independently and simultaneously in parallel.

  The determination unit 18 determines the number included in the recognition target region by adding the calculation result of the color ratio calculation unit 17 to the determination condition of each number. In this way, the numerical recognition accuracy is improved by taking into account the calculation result of the color ratio calculation unit 17. The determination unit 18 can determine that the symbol included in the recognition target area is not a number when the determination condition for each number and the calculation result of the color ratio calculation unit 17 contradict each other. It is also possible to determine that the target binarized image is not a road traffic sign image.

  Of the numbers, two of “0” and “8” do not have distinctive features in the shape of the character, and therefore, completely different images tend to be erroneously recognized as the numbers “0” and “8”. Therefore, in this embodiment, the recognition accuracy of the numbers “0” and “8” can be improved by taking the following special measures.

  FIG. 8 shows the run length characteristics in a partial region with the number “0”. When attention is paid to the partial area BL7 (see FIG. 3) when the number “0” is included in the recognition target area, the black and white columns of a predetermined length or more are identified from the characteristics of the font design of the number “0”. There are many. Therefore, the run length determination unit 162 in the feature amount extraction unit 16_7 to which the image of the partial area BL7 is input determines whether or not a predetermined number or more of black and white columns having a predetermined length or more exist in the partial area BL7. . The predetermined length is an arbitrary length such as 90% or 95% with respect to the height of the partial region BL7, for example. The height of the partial area BL7 can also be specified as a predetermined length. In this case, the run length determination unit 162 determines whether or not there are a predetermined number or more of all-black columns and all-white columns in the partial region BL7. judge. The predetermined number may be an integer of 1 or more, and can be changed according to the size of the scan window.

  The determination unit 18 determines that the number included in the recognition target region is “0” by adding the determination result of the run length determination unit 162 in the feature amount extraction unit 16_7 to the determination condition of the number “0”. can do. Thereby, the recognition accuracy of the number “0” is improved.

  FIG. 9 shows the run length characteristics in several other sub-regions with the number “0”. When attention is paid to the partial areas BL1 to BL3 and BL5 (see FIG. 3) in the case where the recognition target area includes the number “0”, the partial areas BL1 and BL2 have a characteristic of the font design of the number “0”. There are columns having a predetermined length or more composed of pixels of a predetermined color, and partial regions BL3 and BL5 have lines having a predetermined length or more composed of pixels of a predetermined color. Therefore, the run length determination unit 162 in the feature amount extraction units 16_1 and 16_2 to which the images of the partial areas BL1 and BL2 are input has columns of a predetermined length or more composed of pixels of a predetermined color in the partial areas BL1 and BL2, respectively. The run length determination unit 162 in the feature amount extraction units 16_3 and 16_5 to which the images of the partial areas BL3 and BL5 are input is equal to or longer than a predetermined length composed of pixels of a predetermined color in the partial areas BL3 and BL5, respectively. It is determined whether or not a line exists. The predetermined length is, for example, half the height of the partial region for the partial regions BL1 and BL2, and half the width of the partial region for the partial regions BL3 and BL5.

  Further, the determination unit 18 can determine that the number included in the recognition target region is “0” in consideration of the determination results of the run length determination unit 162 in the feature amount extraction units 16_1 to 16_3 and 16_5. Thereby, the recognition accuracy of the number “0” is further improved.

  FIG. 10 shows run length characteristics in several subregions of the number “8”. When attention is paid to the partial areas BL3 to BL5 (see FIG. 3) when the number “8” is included in the recognition target area, the partial areas BL3 to BL5 have a predetermined color due to the characteristics of the font design of the number “8”. There is a line of a predetermined length or more consisting of the pixels. Therefore, the run length determination unit 162 in the feature amount extraction units 16_3 to 16_5 to which the images of the partial areas BL3 to BL5 are input has lines of a predetermined length or more composed of pixels of a predetermined color in the partial areas BL3 to BL5, respectively. It is determined whether or not. The predetermined length is, for example, half of the width of each of the partial regions BL3 to BL5.

  The determination unit 18 adds the determination result of the run length determination unit 162 in the feature amount extraction units 16_3 to 16_5 to the determination condition of the number “8”, and the number included in the recognition target region is “8”. Can be determined. Thereby, the recognition accuracy of the number “8” is improved. Note that the determination result of the run length determination unit 162 in the feature amount extraction units 16_3 and 16_5 can be used in common for the determination processing of the numbers “0” and “8” in the determination unit 18.

  As described above, according to the present embodiment, the accuracy of recognizing numbers in road speed signs can be improved.

<< Third Embodiment >>
FIG. 11 shows the configuration of the main part of a symbol recognition apparatus according to the third embodiment of the present invention. The symbol recognition device 10 according to the present embodiment is obtained by adding feature quantity extraction units 16_1U to 16_11U, 16_1D to 16_11D, 16_1L to 16_11L, and 16_1R to 16_11R to the symbol recognition device 10 according to the first embodiment. Hereinafter, description of the same matters as in the first embodiment will be omitted, and points different from those in the first embodiment will be mainly described.

  If a recognition object such as a vehicle sign is not correctly captured in the scan window, a symbol to be recognized may protrude from the recognition target area and may not be recognized correctly. Therefore, the symbol recognition apparatus 10 according to the present embodiment simultaneously performs a recognition process on a reference recognition target region located substantially in the center of the scan window and each recognition target region slightly shifted vertically and horizontally around the reference recognition target region. I do.

  FIG. 12 shows an example of a reference recognition target area and an example of a recognition target area shifted vertically and horizontally. Images of the partial areas of the reference recognition target area NW are input to the feature amount extraction units 16_1 to 16_11. The images of the partial areas of the recognition target area NW shifted upward are input to the feature amount extraction units 16_1U to 16_11U. The images of the partial areas of the recognition target area NW shifted downward are input to the feature amount extraction units 16_1D to 16_11D. The images of the partial areas of the recognition target area NW shifted to the left are input to the feature amount extraction units 16_1L to 16_11L. The images of the partial areas of the recognition target area NW shifted to the right are input to the feature amount extraction units 16_1R to 16_11R. The feature amount extraction units 16_1U to 16_11U, 16_1D to 16_11D, 16_1L to 16_11L, and 16_1R to 16_11R are elements having the same configuration as the feature amount extraction units 16_1 to 16_11, and input images are input to the feature amount extraction units 16_1 to 16_11. It is only different from the input image.

  The shift amount of the recognition target area is, for example, 1 pixel when one side of the scan window is 20 to 29 pixels, 2 pixels when one side of the scan window is 30 to 39 pixels, and one side of the scan window is In the case of 40 to 50 pixels, it can be appropriately changed according to the size of the scan window, such as 3 pixels.

  The control unit 14 outputs the control signal ACT, and the histogram calculation unit 161 and the run length determination unit 162 in each of the feature amount extraction units 16_1 to 16_11, 16_1U to 16_11U, 16_1D to 16_11D, 16_1L to 16_11L, and 16_1R to 16_11R. Controls the input of pixel information to each. For example, each histogram calculation unit 161 and each run length determination unit 162 fetches the pixel value Ixy and coordinate data (x, y) when the control signal ACT input to the histogram calculation unit 161 and the run length determination unit 162 are at the H level, and the control signal ACT is at the L level. In this case, the pixel value Ixy and the coordinate data (x, y) are not captured. Thereby, the control unit 14 scans the binarized image stored in the image memory 12 once, and the image of an arbitrary partial area of the binarized image is transferred to each histogram calculation unit 161 and each run length determination unit 162. Can be entered. Each histogram calculation unit 161 and each run length determination unit 162 perform a predetermined process on the input image simultaneously and independently from each other.

  The determination unit 18 determines the numbers included in each recognition target region based on the outputs of the feature amount extraction units 16_1 to 16_11, 16_1U to 16_11U, 16_1D to 16_11D, 16_1L to 16_11L, and 16_1R to 16_11R. The determination unit 18 can determine the symbol recognition result by majority decision of the recognition result of each recognition target region.

  As described above, according to the present embodiment, symbols included in the binarized image can be correctly recognized even when the recognition target is not correctly captured in the scan window.

  Also in the symbol recognition apparatus 10 according to the second embodiment, the recognition target object is correctly captured in the scan window by adding the color ratio calculation unit 17 to which the images of the recognition target regions shifted in the vertical and horizontal directions are input. Even if it is not, it is possible to correctly recognize the symbols included in the binarized image.

<< Fourth Embodiment >>
For example, as shown in FIG. 2, a road speed sign represented by a design in which a two-digit number is surrounded by a circle frame is captured in the scan window SW together with the circle frame. The symbol recognition apparatus 10 according to the first to third embodiments performs image analysis by dividing the recognition target region into several partial regions, with the substantially central portion of such a scan window SW as a recognition target region, The scan window SW, that is, the symbol included in the binarized image is recognized. On the other hand, as shown in FIG. 2, in the scan window SW, the area around the recognition target area includes a circle frame of road speed signs. The circle frame can be detected by the above-mentioned method, and it can be determined from the detection result whether the scan window SW, that is, the binarized image is an image of a road speed sign. That is, the above-described symbol recognition device 10 can be expanded to constitute a vehicle sign recognition device.

  FIG. 13: shows the structure of the principal part of the vehicle marker recognition apparatus which concerns on the 4th Embodiment of this invention. The vehicle sign recognition apparatus 100 according to the present embodiment is obtained by adding a frame detection unit 20 and an oblique line determination unit 30 to the symbol recognition apparatus 10 according to the first embodiment. Hereinafter, description of the same matters as in the first embodiment will be omitted, and points different from those in the first embodiment will be mainly described.

  The frame detection unit 20 detects the frame of the vehicle sign in the binarized image stored in the image memory 12. Image information is input to the frame detection unit 20 directly from the image memory 12 or indirectly through the control unit 14. For example, a road speed sign has a circle around a two-digit number as shown in FIG. Therefore, in order to detect such a circular frame, the frame detection unit 20 has a predetermined number of lines, columns, or diagonal lines having pixels of a predetermined color or longer, which are composed of pixels of a predetermined color in some partial areas of the binarized image. It can be determined whether or not it exists, and the frame of the vehicle sign can be detected from the determination result. The predetermined color is black in the case of a normal vehicle sign, and white in the case of an electric display sign.

  FIG. 14 shows an example of a partial area of the binarized image processed by the frame detection unit 20. The frame detection unit 20 has a predetermined number or more of lines, columns, or diagonal lines of pixels of a predetermined color in each of the vertical, horizontal and diagonal eight partial areas BL21 to BL28 in the scan window SW. It is determined whether or not to do. Specifically, the frame detection unit 20 determines whether or not there are a predetermined number or more columns of pixels having a predetermined color in the partial areas BL21 and BL23, and the predetermined colors of the partial areas BL22 and BL24. It is determined whether or not there are a predetermined number or more of pixels having a predetermined length or more, and in the partial areas BL25 and BL27, a predetermined number of 45 ° diagonal lines extending from the upper left to the lower right and having a predetermined length or more of pixels of a predetermined color. It is determined whether or not there is more than the predetermined number, and it is determined whether or not there are a predetermined number or more of 45 degree diagonal lines from the lower left to the upper right, each having a predetermined length or more, in the partial areas BL26 and BL28. The predetermined length is an arbitrary length such as 90% or 95% with respect to the length in the longitudinal direction of each partial region. The predetermined number is an arbitrary number such as one. When the frame detection unit 20 determines that there are a predetermined number or more of lines, columns, or diagonal lines having a predetermined length or more of pixels of a predetermined color in all the partial areas BL21 to BL28, the road speed in the binarized image It is determined that there is a circle frame used for a sign.

  Note that the frame of the vehicle sign detected by the frame detection unit 20 is not limited to a circular frame. By appropriately changing the partial area to be processed by the frame detection unit 20, it is possible to correspond to detection of a diamond frame used as a warning sign, an inverted triangle frame used as a slow sign, a stop sign, or the like. it can.

  By the way, in addition to road speed signs, there are signs indicating prohibition such as U-turn prohibition and overtaking prohibition (hereinafter referred to as “prohibition signs”), and these prohibition signs have a diagonal line indicating prohibition in common. It is included. Returning to FIG. 13, the oblique line determination unit 30 determines whether or not such an oblique line exists in the binarized image stored in the image memory 12. Specifically, the oblique line determination unit 30 determines whether or not there is an oblique line having a predetermined length or more, which is composed of pixels of a predetermined color, in an oblique partial area in the binarized image stored in the image memory 12. Image information is input to the oblique line determination unit 30 directly from the image memory 12 or indirectly through the control unit 14. The predetermined color is black in the case of a normal vehicle sign, and white in the case of an electric display sign.

  FIG. 15 shows examples of partial regions of the binarized image processed by the oblique line determination unit 30 and some examples of signs indicating prohibition. The oblique line determination unit 30 determines whether or not there is an oblique line having a predetermined length or more made of pixels of a predetermined color in the oblique rectangular partial region BL29 in the scan window SW.

  Returning to FIG. 13, the control unit 14 outputs a control signal ACT to control input of pixel information to the frame detection unit 20 and the oblique line determination unit 30. For example, the frame detection unit 20 and the oblique line determination unit 30 take in the pixel value Ixy and coordinate data (x, y) when the control signal ACT input to the frame detection unit 20 and the oblique line determination unit 30 are at the H level, and when the control signal ACT is at the L level. Pixel value Ixy and coordinate data (x, y) are not captured. Thus, the control unit 14 scans the binarized image stored in the image memory 12 once, and the image of an arbitrary partial region of the binarized image is extracted from the feature amount extraction units 16_1 to 16_11, the frame detection unit 20, And can be input to the oblique line determination unit 30. Then, each of the feature amount extraction units 16_1 to 16_11, the frame detection unit 20, and the oblique line determination unit 30 performs a predetermined process on the input image independently and simultaneously in parallel.

  The determination unit 18 receives the outputs of the feature amount extraction units 16_1 to 16_11, the frame detection unit 20, and the oblique line determination unit 30, and determines whether or not the binarized image stored in the image memory 12 is a vehicle sign. If it is a vehicle sign, it is a prohibited sign, if it is a prohibited sign, what kind of prohibited sign it is, and if it is a speed sign instead of a prohibited sign, its speed limit Recognize First, based on the detection result of the frame detection unit 20, the determination unit 18 determines that the binarized image to be recognized is an image of the vehicle sign when the binarized image does not include the frame of the vehicle sign. Judge that there is no. When the frame of the vehicle sign is detected by the frame detection unit 20, the determination unit 18 determines that the binarized image includes a diagonal line indicating prohibition based on the determination result of the diagonal line determination unit 30. It is determined that the binarized image to be recognized is a prohibition sign image. Further, when the frame detection unit 20 detects the circle frame of the road speed sign and the diagonal line determination unit 30 determines that there is no diagonal line representing prohibition in the binarized image, the determination unit 18 performs binary processing. The symbols included in the digitized image are recognized as numbers representing the speed limit.

  When the determination unit 18 determines that the binarized image to be recognized is an image of a prohibition sign, the determination unit 18 further outputs the output of the feature amount extraction units 16_1 to 16_11, that is, the histogram and run length of each partial region of the recognition target region. The type of prohibition sign can also be determined based on the above.

  In some cases, it is better to use the histogram and run length of the image of the recognition target area and the partial area suitable for the determination of the type of the prohibited sign. Therefore, in such a case, a feature amount extraction unit to which pixel information of a partial area for determining the type of the prohibited sign is input is provided separately, and the determination unit 18 is based on the output of the feature amount extraction unit. The type of prohibition sign may be determined.

  As described above, according to the present embodiment, determination of vehicle signs and determination of prohibition signs can be performed only by scanning the binary image stored in the image memory 12 once without repeatedly reading the binary image from the image memory 12. It is possible to recognize the type of prohibition sign and the speed limit. Therefore, the vehicle sign recognition apparatus 100 according to the present embodiment is suitable for mounting on hardware such as an FPGA (Field Programmable Gate Array). In addition, by mounting the vehicle sign recognition apparatus 100 according to the present embodiment on an FPGA, it is possible to realize a low-cost in-vehicle built-in system that can perform vehicle sign recognition processing in real time while the automobile is running. it can.

  Also for the symbol recognition device 10 according to the second and third embodiments, a vehicle sign recognition device similar to the above can be configured by adding the frame detection unit 20 and the oblique line determination unit 30.

DESCRIPTION OF SYMBOLS 10 Symbol recognition apparatus 12 Image memory 14 Control part 161 Histogram calculation part 162 Run length determination part 18 Determination part 20 Frame detection part 30 Diagonal line determination part 100 Sign recognition apparatus for vehicles

Claims (9)

An image memory for storing a binarized image to be recognized;
A plurality of histogram calculation units for inputting an image of each partial area of the recognition target area in the binarized image and calculating a frequency distribution of pixels of a predetermined color in each line or each column in the partial area;
A plurality of run-length determination units that input an image of each partial area of the recognition target area and determine whether or not a line or a column having a predetermined length or more composed of pixels of the predetermined color exists in the partial area;
A control unit for inputting pixel information of the partial area read by scanning the binarized image stored in the image memory to the plurality of histogram calculation units and the plurality of run length determination units;
A symbol recognition unit, comprising: a determination unit configured to determine a symbol included in the binarized image based on a calculation result of the plurality of histogram calculation units and a determination result of the plurality of run length determination units. A device ,
The plurality of histogram calculation units divide the recognition target region into two parts in the left-right direction, first and second histogram calculation units to which an image of each partial region is input, and the recognition target region is divided into three in the vertical direction. Third, fourth, and fifth histogram calculators to which images of each partial area are input; a sixth histogram calculator to which an image of a partial area at the center in the horizontal direction of the recognition target area; A seventh histogram calculation unit to which an image of a partial area at the center in the vertical direction of the recognition target area is input,
The first, second, third, and fourth runs in which the plurality of run length determination units are inputted with images of the partial areas obtained by dividing the partial area at the center in the vertical direction of the recognition target area into four parts in the vertical and horizontal directions. Including a length determination unit,
The determination unit includes a maximum value of the histogram calculated by the first to fifth histogram calculation units, a minimum value of the histogram calculated by the sixth and seventh histogram calculation units, and the first to fourth values. The symbols included in the recognition target area are determined based on the determination result of the run length determination unit
Symbol recognition device . The plurality of run length determination units includes a fifth run length determination unit,
Whether the fifth run length determination unit receives an image of a partial area in the center in the vertical direction of the recognition target area, and whether or not there are a predetermined number or more of black and white columns having a predetermined length or longer in the partial area. Is determined,
The determination unit, by adding the determination result of the fifth run length determination unit, according to claim 1 in which the symbol included in the recognition target region and judging whether or not the number "0" The symbol recognition device described in 1. The plurality of run length determination units include sixth and seventh run length determination units to which an image of each partial region obtained by dividing the recognition target region into two in the left-right direction, and three recognition target regions in the vertical direction. An eighth and a ninth run length determination unit to which images of the divided upper and lower partial areas are input,
The determination unit is configured to determine whether or not a symbol included in the recognition target area is a number “0” in consideration of the determination results of the sixth to ninth run length determination units. The symbol recognition device according to claim 1 or 2 . The plurality of run length determination units include sixth, seventh, and eighth run length determination units to which images of partial regions obtained by dividing the recognition target region into three in the vertical direction are input,
The determination unit is configured to determine whether or not the symbol included in the recognition target region is a number “8” in consideration of the determination results of the sixth to eighth run length determination units. The symbol recognition device according to claim 1 or 2 . The plurality of run length determination units includes a tenth run length determination unit to which an image of a middle partial region obtained by dividing the recognition target region into three in the vertical direction is input,
The determination unit is configured to determine whether or not the symbol included in the recognition target region is a number “8” in consideration of the determination results of the eighth to tenth run length determination units. The symbol recognition device according to claim 3 . An image of the recognition target area is input, and a color ratio calculation unit that calculates a ratio of pixels of a predetermined color in the entire recognition target area is provided.
The control unit inputs pixel information of the partial area read by scanning the binarized image stored in the image memory to the color ratio calculation unit,
The determination unit, in consideration of the calculation result of the color ratio calculating section, a symbol recognition device according to any one of claims 1 to 5, characterized in that to determine the symbols included in the recognition target region . The symbol recognition according to any one of claims 1 to 6 , wherein recognition processing is performed in parallel for each recognition target region in which the recognition target region and the recognition target region are slightly shifted vertically and horizontally. apparatus. A symbol recognition device according to any one of claims 1 to 7 ,
A frame detector for detecting a frame of a vehicle sign in the binarized image stored in the image memory of the symbol recognition device;
The vehicle sign recognition apparatus, wherein the determination unit of the symbol recognition apparatus determines whether the binarized image is an image of a vehicle sign based on a detection result of the frame detection unit. . An oblique line determination unit for determining whether or not there is an oblique line having a predetermined length or more composed of pixels of a predetermined color in an oblique partial region in the binarized image;
The determination unit, in consideration of a determination result of the oblique line determination unit, determines which type of sign of the sign indicating prohibition the binarized image is. vehicular marker recognition apparatus according to 8.